The Future of Verifiability: Exploring ZK Coprocessors

GM, GM. Welcome to what's going to be another great episode of DevNTell. So, if you didn't know, DevNTell is a 30-minute podcast held every week, allowing founders, hackers, and anyone in between to come on the show and give an explanation of their product. So, today I'm ecstatic to welcome Aryan Sheikhalian, who's head of research at CMT Digital. If you didn't know, CMT Digital is a leading global venture capital firm focused on early-stage investments that accelerate the adoption of blockchain technology. So, if you stick around for today's episode, you'll see Aryan give us an engaging talk exploring zero-knowledge coprocessors, zero-knowledge in general, and their role in the future of verifiability. All right, let's get going.

GM, GM. Welcome to the show, Aryan. A pleasure to have you on.

Thank you for having me.

Yeah, my pleasure, my pleasure. Ah, so, we're going to have a really great, engaging talk today all about zero-knowledge and zero-knowledge coprocessors. This is a big, big topic in the Web3 space, as we know, but I'm pretty sure there is quite a bit of applications beyond just Web3 and really keen to hear the insights that you have here for us today.

But before we get into that, let's, let's dive in a little bit and learn more about yourself. So, would you like to give a brief intro, just for folks who aren't familiar?

Yeah, for sure. My name is Aryan. I've been with CMT for about four years. I work as the head of research here and as an investor. I was born in Iran, moved to the UK when I was around four, grew up in the UK, and I came to the United States for university back in 2016. Studied mathematics and economics at Columbia and joined CMT when I graduated. Just got my visa, so looks like I'll be in the United States for at least the immediate future, and yeah, I'm excited for the opportunities that this country provides.

Awesome, man. Congratulations. I'm sure the visa process, I've heard, is never an easy one. So, you're getting through that. That's amazing.

I guess, before we get into the nitty-gritty of zero-knowledge, you were telling me a great, your background story, basically, of how you got into tech and how you found yourself at CMT. Would you like to share that? I thought it was a really great story.

Yeah, for sure. I mean, I was always very interested in technology. I think it started when I was a kid. In the UK, we've got these things called car boot sales. It's kind of like a yard sale where a bunch of cars will drive into a field and pop the trunk and sell various household items from their bag. I was really into computers and so I would go to these car boot sales and find old computers that could be pieced together and sold and that's how I first got into technology was really just understanding how computers worked. Obviously computers have come a long way since then. Back then I was actually purchasing like individual networking cards, audio cards, all of that is now kind of integrated in the motherboard. So, I, when I finished high school, I started working at Accenture. Accenture runs a program called the Horizon Scheme where they let a few gap year students work as normal business analysts. So, I was drawn to Accenture because of its focus on technology and the actual implementation side of the solution. That's where I first got exposed to blockchain technology and found it really interesting, started diving in and went from there.

Awesome. And I guess, is there a particular thing with blockchain technology that kind of hooked you? I know for a lot of folks, their first dive into it is reading the Bitcoin white paper and kind of getting hooked from that. Was there a particular thing that you researched or learned about that really hooked you into the tech?

Of course the Bitcoin white paper was hugely influential for kind of my interest in the space. I read the Bitcoin white paper before I joined Accenture. What about my Accenture experience really interested me in the technology was I was exposed to a lot of inefficiencies that exist within the modern world, especially the consulting world, especially technology that exists within enterprises and the automation abilities of blockchain technology in handling some of those processes was really interesting to me. So, that's kind of the angle from which I approached it: how can we use blockchain technology to automate a lot of these inefficiencies away and create a better system than exists today?

Oh, yeah, definitely. That is, that is a great way to get hooked.

And you mentioned you're head of research at CMT Digital. Could you perhaps just give us a high-level overview of the types of different things that CMT Digital invests in and, like, more about your role?

Yeah, for sure. So, I'll tell you a little bit about CMT Digital as an organization as well. So, CMT Digital really exists as the blockchain investment subsidiary of the CMT Group and the CMT Group is a proprietary trading firm based here in Chicago that's been around since 1997. We, the firm was actually established around the genesis of electronic trading and our founders Scott and Jan came out of Chicago research and trading. They saw the opportunities in the electronic markets and kind of set up the firm, set up the firm to engage in those markets. CMT Digital was established in 2013 to start engaging in the blockchain and digital asset space, primarily at the beginning it was from a regulatory and policy perspective. You know, we observed the Silk Road auction and were wondering about the ramifications of the US Marshal Service selling something like Bitcoin, an asset that I don't think many people understood at that time. And so we joined the Blockchain Association, made initial investments in 2015, the crypto trading desk was spun up full-time in 2017, and the first venture fund was launched in 2018. I joined around the second fund in 2021 and, yeah, we're on our fourth fund right now. For my role, you know, a head of research role is really a bit of a hybrid of an investor and a researcher role, especially at a venture capital firm. You're looking at the research side to really inform the investing strategies to a degree. So, being able to dive into the research and develop theses about what verticals in the market we expect disruption in or how that disruption is going to look, I think informs our investing strategy. And I think that's something that we experience with ZK as well. For example, we had a thesis about limited use of ZK for computation and that's how we ultimately arrived at investing in Lagrange at the pre-seed level. So, yeah, it's definitely something where the research informs the investing and vice versa.

Interesting, interesting. And I guess, could you take us through a day in the life of your role? Like, you wake up in the morning, like, what's the first thing you do? Is it just like scour the web for the latest research papers or?

Well, I think it's a mixture. So we're obviously seeing deal flow, but the deal flow also informs I guess where people are spending time. And through that we can see if there are areas that we think are underdeveloped and that is probably where the research will be focused. This is also especially prevalent in times of quiet market activity. If there's low deal flow volumes, for example, it gives me an opportunity to really dive into the research aspect and build out the theses further to anticipate builders returning. And so a typical day in the life is I tend to get to the office around 7:30, 8:00 AM, log on, we go through the deals that we're seeing and if there are any research projects going on at the time, we'll be able to explore them as well. One that I'm currently working on is around the tragedy of the commons and how crypto networks and blockchain incentives specifically can alleviate some of the misalignments of incentives that occur as a result of the tragedy of the commons. So, depending on the amount of deal flow we're getting, we can allocate more or less time to the research activities.

Interesting, interesting. Yeah, I've always been curious to learn what a typical day in the VC world is. So, thank you for enlightening us on that. And obviously, zero-knowledge is something you're passionate about, it's the topic of our talk today. So, let's get into that.

So, for people that have never heard of zero-knowledge before, could you just give us a brief rundown of what it is and why we should care about it?

Yeah, zero-knowledge is really a mathematical concept and it's prevalent in cryptography, which is a branch of mathematics. It's a technique where one party can prove to another some statement is true or false without revealing any information about that statement beyond the validity of the statement itself. So, in blockchain, it's often used to prove computations happened correctly or by a specific spec without actually exposing the underlying data.

And this is a recent discovery, right? That this type of thing is possible. And I believe the biggest application, at least that I've seen, has been in the, in the Web3 space, but is there any other field that's adopting zero-knowledge?

I mean, zero-knowledge I think has been prevalent in computer systems for a long time. It's actually not necessarily unique to crypto at all. It's just within crypto there are applications of it that I think become very useful because of the nature of crypto systems and the transparent nature of crypto systems specifically. But zero-knowledge has been a technique used in cryptography for a long time, especially when it comes to secret sharing and establishing the grounds for a communication, for example.

Gotcha, gotcha. And obviously, like privacy and verification are also important aspects that zero-knowledge provides. Like, what do you think they're so important for the different types of tech that we're seeing today and perhaps the tech that we'll see in the future? How do you see all it kind of like coming along?

Yeah, I think ZK is really important because it allows you to decouple trust from transparency. So instead of having to reveal all of your data to prove something, like your credit score or your age, your identity even, you can just prove the result. And so as we move towards a world with increasing surveillance and data exploitation for verification, ZK brings that math-based trust that allows you to scale as those systems scale and evolve and change and verify the correctness associated with your data without actually re-executing computations or revealing privacy-private data. And so this kind of two aspects: it's huge for scaling blockchains, improving off-chain computations, and it's also great for preserving privacy of data that's sensitive, especially in an environment where everything is transparent and verifiable.

Yeah, indeed, indeed. And just to shift gears just a little bit from blockchain, like what about artificial intelligence? We see a lot of data pumping these models and training them and sometimes, well most of the time, it's just like open data that's on the web that scrapes, that they scrape and they train on. Have you seen any developments in basically verifying the output an AI model might produce, or any interesting developments in the AI space related to ZK?

Yeah, I think AI is definitely a branch of ZK. We've got ZKML, so zero-knowledge machine learning, and portfolio company Lagrange is actually working on a ZKML library as well. So, there's the ZKEVM side, so projects like Scroll, Polygon ZKEVM, ZKSync, they're bringing ZK to Ethereum-compatible smart contracts. That's one kind of branch. We've got ZKML, so zero-knowledge machine learning, where we're seeing early examples of proving model inference in zero-knowledge. This verifies that the AI model made a decision fairly without revealing kind of the model or the data inputs. And I would say there's also another branch that's like recursive proofs and proof aggregation. And this is also where Lagrange kind of operates. Innovations here talk about drastically reducing cost and allowing computation proofs being composable and kind of aggregating these proofs to save on computation costs. And this is key for scaling. So ZKML I think is exciting for just showcasing the ability for ZK to quickly be able to solve problems across industries as a result of its inherent properties regardless of its use case.

Interesting, interesting. And I guess the topic of today's show is exploring ZK coprocessors. Could you explain what that is and why you think that's important?

Yeah, so ZK coprocessors are like outsourced verifiable compute. So instead of having all of the computation on-chain, which is very expensive, sometimes not even feasible given the computation size, you send it off-chain to a ZK coprocessor. This is some computation that can occur off-chain, but a succinct proof is then provided that the result is valid, that the computation occurred in some fashion. And why do I think this is important? Well, one of the things that's really exciting about ZK is it's becoming programmable. It's moving from this research phase into this real developer tooling phase. And so using ZK coprocessors in your stack I think allows you to unlock things like scalability for sure, but then private and verifiable compute to unlock private voting, verifiable off-chain inference, cross-chain asset bridges with no trust assumptions, all of them, everything that could be thought of as like a verifiable oracle but for processes like compute can then exist in a scalable and economically viable way today as kind of like verifiable oracles.

Ah, interesting. So ZK as an oracle, essentially. Yeah, never, never would have thought I'd connect those two, but that's an interesting way to, to look at it.

And I guess, are, would you say that coprocessors are in themselves like a specialized set of hardware to be able to run that level of computation? So we know, like, when we run AI models, you need a pretty beefy, beefy computer with lots of RAM and GPU. Is there specialized hardware that's necessary to run the, the zero-knowledge coprocessors as well?

I don't know if it necessarily needs to be specialized hardware. Of course, people are working on hardware acceleration for kind of making compute more efficient on the hardware. Coprocessors are not hardware per se, but more like verifiable compute modules. Typically they're software systems that could be accelerated by that hardware acceleration, but it's about the circuit that's created that verifies the computation that's actually occurring. So the coprocessor piece is really software, but the, they can be accelerated by specialized hardware and we are seeing companies building in the specialized hardware space.

I'm sure, I'm sure they are. And, like, in your mind, do you see, like how do you see these coprocessors playing into the future of verifiability? I know there's lots of applications, but is there something in particular that you're quite keen on seeing being developed?

I think it's about the notion of changing how we think about trust in computation. You no longer then need to rerun or trust the machine, there's no trust between the machine and the protocol, it's really that trust is obtained by verifying the proof and the proof's validity. And so it can help supercharge modular blockchains, things like settlement and execution layers make them more compatible with a wider array of use cases as a result. It can enable regulated privacy, so you know we love privacy in this space, but we also have to be compliant in this space and so that has been a historically difficult bridge to cross. How do you merge those two? It's kind of like an oxymoron, but with this technology it does really enable regulated privacy. You can just follow the rules without showing the data. And, yeah, I think the AI specifically, ZKML does make AI and blockchain interoperable in that trustless way and scalable way, and if you think about the role of blockchain within AI, the incentive alignment abilities of blockchains can be quite useful for AI and it really requires that link of being able to not only verify the processes of AI but then trust it to ensure that you can actually align the incentives and create an incentive model that is not broken as a result of broken inputs.

Right, right. I personally think that's also going to be quite important going forward, especially, like this, this year in particular and around the end of last year, you really saw this merging of crypto and or blockchains and AI really starting to take off with like the Eliza Agent stuff. Of course not everything's going to be automated meme coin reply guy bots, of course, but I think like going forward there's going to be lots of interesting applications that kind of come out of this, this merge. And obviously you being in venture capital and CMT having quite the big portfolio, I'm sure there's quite a bit of companies that you have your eye on or do have inside your portfolio. Lagrange being one that you mentioned. Are there any other ones that are doing interesting work in the ZK space or maybe the blockchain and AI space?

Yeah, I mean I think there's a lot of companies in the, in the space that are interesting and increasingly so more and more companies are finding a need for ZK in their stack. I am quite excited about ZKTLS, so companies like Reclaim and Opacity are working on using the transport layer in internet protocols today like TLS, transport layer security, to prove data that exists in this backlog of Web2 data lakes. And so all of your information related to, you know, your university where you went to school, your employment, your credit, all of that is obtainable via internet protocols. You can log into your bank account, for example. But using ZKTLS to actually prove the inputs to the page that you're seeing, you can then verify some information is true without actually revealing that information. And I think that's powerful for similar to AI onboarding data and making that composable with blockchains, therefore allowing blockchains to shine where they're really best, which is that incentive alignment piece using data that is then verifiable and known to be accurate.

Yeah, definitely. And I think another interesting like on that topic, an application of this would be, have you heard of, have you heard of Clear before? The airport terminals that like scan... Yeah. I remember. Oh, you remember, yeah. There you go, you know, you know the whole deal. I feel that like something like that would benefit a lot from ZK just because of like the amount of information that you have to give. And then there'd be also interesting applications of that where you could also technically use that for, like, sign- or checking into your hotel as well or checking into a cruise, just because like all of that is kind of housed in one central identity proving platform and you don't technically have to prove- you don't have to provide them all your information if you're using ZK- something like ZKTLS or some other zero-knowledge application. But...

Yeah, one example that I think is relevant today is the 23andMe database. So the database is going up for sale, a judge ruled that it can be sold as part of the bankruptcy proceedings. I don't know if you saw but Sei, the blockchain SEI, is submitting a bid. Now what's interesting is a blockchain company is uniquely positioned to buy the database, make it all zero-knowledge, give people who are within that database access to their credentials, and then build a system for identity as a result of having access to that data without ever compromising the security or privacy of that database itself. And so it alleviates the privacy concerns of the users whilst also giving them access to an identity proof that they can then use in blockchain systems. Just a question of whether or not they can win the bid.

Yeah, that would be, that would be an interesting development if they do end up winning that bid. And yeah, I mean, 23andMe has quite the large database and it's like going to the right hands I think is very important for all the people that gave up their data basically to get their DNA test. But yeah, I guess on that front, as we're nearing time, just wanted to pick your brain a little bit.

As you're right on the bleeding edge researching all the different types of innovations in Web3 and AI, zero-knowledge, are there any trends that you can speak on that you've you see are developing or you're quite interested in yourself, like starting to come up?

Well, you know, I think a lot of the past few years have been spent especially in the blockchain space developing infrastructure. And you know, I don't think we'll ever be fully finished developing infrastructure, but at least the rate at which infrastructure is being made and applications are being made I think should be changing soon. I think we've hit an inflection point where the infrastructure is quite robust broadly. And tools like ZKTLS specifically and recursive proofs especially can enable some types of consumer applications that we couldn't before. And so I am quite excited about seeing consumer applications using blockchain technology and the suite of tools like ZKTLS that accompany it. And so I think it's a matter of seeing where blockchain can actually disrupt: where does that self-ownership and incentive alignment piece within a broader network actually disrupt incumbents? And I think applications of that are pretty interesting. Obviously identity, data are pretty low-hanging fruit, but as we see the TradFi and DeFi world merge, I'm excited to see what sort of consumer applications come out of that as a result.

Yeah, yeah, likewise. The biggest thing, at least from the blockchain standpoint, as it's been quite the theme ever since I've come into the Web3 space is just like making something that's usable or not making it obvious that what you're using is like blockchain-enabled or ZK-enabled. Basically abstracting that, all that fancy stuff to the back and just making usable apps for people to use.

And I think that's where blockchain's promise really lies. Like, we can't beat Web2 UX by just copying it because the blockchain stack looks fundamentally different. But there are applications or places where the blockchain stack is going to be better than the Web2 stack. And I think things like privacy without compromise on using ZK is a great example of that. You know, being able to prove that you're over 18 without revealing your actual birthday. You know, healthcare, share that you've taken a COVID test, for example, without sharing that personal medical data. There's a lot of things that are very much harder to do in the Web2 world than the Web3 world, assuming the Web3 world is able to battle-test and adopt the technology to the point where it's accepted by users as being trustable. And I think there is a little bit of not education but kind of getting used to using mathematical proofs. You know, we use a lot of cryptographic elements in our everyday lives when we're logging into web portals or whatever, and I think it's going to be a similar sort of adoption curve where we just need to get used to the verifiability and trust that comes from ZK systems broadly.

Well said, well said. And yeah, that brings us to time. Yeah, this was a wonderful, wonderful talk Aryan. Thank you so much for taking the time out of your day to come on the show today.

Thank you so much for having me. Great to meet you.

Yeah likewise. And for folks that have been watching us live, I'm going to drop the link for you to be able to claim your attendance badge for watching us live today. So if you're into collectibles, definitely check out BitBadges and you can claim your collectible for watching us live today. And with that, I just wanted to wish everybody a very happy Friday, happy weekend wherever you may be, and we will catch you here next week for another great episode of DevNTell. All right y'all, cheers!

Thank you.